X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/5cf8c18e735bb15da3f131e7ae57bb4b222bb4ed..3f8f2a9b9d1ccb1e4e696607ef015de0524da519:/services/crunch-dispatch-slurm/squeue.go?ds=sidebyside

diff --git a/services/crunch-dispatch-slurm/squeue.go b/services/crunch-dispatch-slurm/squeue.go
index e157469711..ee87a44fff 100644
--- a/services/crunch-dispatch-slurm/squeue.go
+++ b/services/crunch-dispatch-slurm/squeue.go
@@ -1,130 +1,108 @@
 package main
 
 import (
-	"bufio"
+	"bytes"
 	"log"
 	"os/exec"
+	"strings"
 	"sync"
 	"time"
 )
 
 // Squeue implements asynchronous polling monitor of the SLURM queue using the
 // command 'squeue'.
-type Squeue struct {
-	squeueContents []string
-	squeueDone     chan struct{}
-	squeueError    error
-	squeueCond     *sync.Cond
-	SlurmLock      sync.Mutex
+type SqueueChecker struct {
+	Period    time.Duration
+	uuids     map[string]bool
+	startOnce sync.Once
+	done      chan struct{}
+	sync.Cond
 }
 
-// squeueFunc
 func squeueFunc() *exec.Cmd {
-	return exec.Command("squeue", "--format=%j")
+	return exec.Command("squeue", "--all", "--format=%j")
 }
 
 var squeueCmd = squeueFunc
 
-// RunSqueue runs squeue once and captures the output.  If there is an error,
-// set "squeueError".  If it succeeds, set "squeueContents" and then wake up
-// any goroutines waiting squeueCond in CheckSqueue().
-func (squeue *Squeue) RunSqueue() error {
-	var newSqueueContents []string
+// HasUUID checks if a given container UUID is in the slurm queue.
+// This does not run squeue directly, but instead blocks until woken
+// up by next successful update of squeue.
+func (sqc *SqueueChecker) HasUUID(uuid string) bool {
+	sqc.startOnce.Do(sqc.start)
 
+	sqc.L.Lock()
+	defer sqc.L.Unlock()
+
+	// block until next squeue broadcast signaling an update.
+	sqc.Wait()
+	return sqc.uuids[uuid]
+}
+
+// Stop stops the squeue monitoring goroutine. Do not call HasUUID
+// after calling Stop.
+func (sqc *SqueueChecker) Stop() {
+	if sqc.done != nil {
+		close(sqc.done)
+	}
+}
+
+// check gets the names of jobs in the SLURM queue (running and
+// queued). If it succeeds, it updates squeue.uuids and wakes up any
+// goroutines that are waiting in HasUUID() or All().
+func (sqc *SqueueChecker) check() {
 	// Mutex between squeue sync and running sbatch or scancel.  This
 	// establishes a sequence so that squeue doesn't run concurrently with
 	// sbatch or scancel; the next update of squeue will occur only after
 	// sbatch or scancel has completed.
-	squeueUpdater.SlurmLock.Lock()
-	defer squeueUpdater.SlurmLock.Unlock()
-
-	// Also ensure unlock on all return paths
-	defer squeueUpdater.squeueCond.L.Unlock()
+	sqc.L.Lock()
+	defer sqc.L.Unlock()
 
 	cmd := squeueCmd()
-	sq, err := cmd.StdoutPipe()
-	if err != nil {
-		log.Printf("Error creating stdout pipe for squeue: %v", err)
-		squeueUpdater.squeueCond.L.Lock()
-		squeueUpdater.squeueError = err
-		return err
-	}
-	cmd.Start()
-	scanner := bufio.NewScanner(sq)
-	for scanner.Scan() {
-		newSqueueContents = append(newSqueueContents, scanner.Text())
-	}
-	if err := scanner.Err(); err != nil {
-		cmd.Wait()
-		log.Printf("Error reading from squeue pipe: %v", err)
-		squeueUpdater.squeueCond.L.Lock()
-		squeueUpdater.squeueError = err
-		return err
+	stdout, stderr := &bytes.Buffer{}, &bytes.Buffer{}
+	cmd.Stdout, cmd.Stderr = stdout, stderr
+	if err := cmd.Run(); err != nil {
+		log.Printf("Error running %q %q: %s %q", cmd.Path, cmd.Args, err, stderr.String())
+		return
 	}
 
-	err = cmd.Wait()
-	if err != nil {
-		log.Printf("Error running squeue: %v", err)
-		squeueUpdater.squeueCond.L.Lock()
-		squeueUpdater.squeueError = err
-		return err
+	uuids := strings.Split(stdout.String(), "\n")
+	sqc.uuids = make(map[string]bool, len(uuids))
+	for _, uuid := range uuids {
+		sqc.uuids[uuid] = true
 	}
-
-	squeueUpdater.squeueCond.L.Lock()
-	squeueUpdater.squeueError = nil
-	squeueUpdater.squeueContents = newSqueueContents
-	squeueUpdater.squeueCond.Broadcast()
-
-	return nil
+	sqc.Broadcast()
 }
 
-// CheckSqueue checks if a given container UUID is in the slurm queue.  This
-// does not run squeue directly, but instead blocks until woken up by next
-// successful update of squeue.
-func (squeue *Squeue) CheckSqueue(uuid string) (bool, error) {
-	squeueUpdater.squeueCond.L.Lock()
-	// block until next squeue broadcast signaling an update.
-	squeueUpdater.squeueCond.Wait()
-	if squeueUpdater.squeueError != nil {
-		e := squeueUpdater.squeueError
-		squeueUpdater.squeueCond.L.Unlock()
-		return false, e
-	}
-	contents := squeueUpdater.squeueContents
-	squeueUpdater.squeueCond.L.Unlock()
-
-	for _, k := range contents {
-		if k == uuid {
-			return true, nil
+// Initialize, and start a goroutine to call check() once per
+// squeue.Period until terminated by calling Stop().
+func (sqc *SqueueChecker) start() {
+	sqc.L = &sync.Mutex{}
+	sqc.done = make(chan struct{})
+	go func() {
+		ticker := time.NewTicker(sqc.Period)
+		for {
+			select {
+			case <-sqc.done:
+				ticker.Stop()
+				return
+			case <-ticker.C:
+				sqc.check()
+			}
 		}
-	}
-	return false, nil
+	}()
 }
 
-// StartMonitor starts the squeue monitoring goroutine.
-func (squeue *Squeue) StartMonitor(pollInterval time.Duration) {
-	squeueUpdater.squeueCond = sync.NewCond(&sync.Mutex{})
-	squeueUpdater.squeueDone = make(chan struct{})
-	squeueUpdater.RunSqueue()
-	go squeueUpdater.SyncSqueue(pollInterval)
-}
-
-// Done stops the squeue monitoring goroutine.
-func (squeue *Squeue) Done() {
-	squeueUpdater.squeueDone <- struct{}{}
-	close(squeueUpdater.squeueDone)
-}
-
-// SyncSqueue periodically polls RunSqueue() at the given duration until
-// terminated by calling Done().
-func (squeue *Squeue) SyncSqueue(pollInterval time.Duration) {
-	ticker := time.NewTicker(pollInterval)
-	for {
-		select {
-		case <-squeueUpdater.squeueDone:
-			return
-		case <-ticker.C:
-			squeueUpdater.RunSqueue()
-		}
+// All waits for the next squeue invocation, and returns all job
+// names reported by squeue.
+func (sqc *SqueueChecker) All() []string {
+	sqc.startOnce.Do(sqc.start)
+	sqc.L.Lock()
+	defer sqc.L.Unlock()
+	sqc.Wait()
+	var uuids []string
+	for uuid := range sqc.uuids {
+		uuids = append(uuids, uuid)
 	}
+	return uuids
 }